Inter-digit HMM: connected digit recognition using the Macrophone corpus

Continuous digit recognition over the telephone channel is a key technology for many telecommunications applications such as voice dialing, automatic banking, and credit card number entry. Speech recognizers usually achieve high performance by modeling the acoustics in hidden Markov models (HMMs) using large numbers of multivariate Gaussian mixtures with assumed diagonal covariance in order to model the variability of different speakers and channel conditions. We present a system that uses single mixture 16 feature Gaussian distribution with an assumed identity covariance to achieve a 1.0% word error and a 5.7% sentence error rate on the Macrophone corpus. We found that inter-digit modeling, discriminant training, and per-utterance adaptation can each contribute about a 30% reduction in the error rate. Using this approach, we can realize a system with relatively low memory requirements.